Skip to content
agentscamp
Term · Term

Design Tokens

Design tokens are named values for a design system's decisions — color, spacing, type, radius — stored once and referenced everywhere instead of hardcoded.

Updated Sep 10, 2026
designdesign-systemtokensconsistency

Design tokens are the named values behind a design system — colors, spacing, type scale, radii, shadows — defined once and referenced by every component. They are what makes a design system legible to an AI agent, which is why token hygiene now affects the quality of generated UI.

Design tokens are the named values that encode a design system's decisions — color, spacing, type scale, radius, shadow, motion — defined in one place and referenced by every component instead of being retyped as raw values. --space-4 instead of 16px, --color-surface-raised instead of a hex code.

The old argument for tokens was consistency and theming: change one value, and every screen follows. The 2026 argument is machine legibility. Every generative tool that produces an interface — Claude Design reading a repository, a prompt-to-prototype tool reading a package — reproduces the structure it finds. Give it a named scale and it reuses your scale; give it eleven similar grays and it invents a twelfth. Token hygiene is now the highest-leverage design work in a codebase, because it compounds through every generated screen after it.

That makes extraction the practical starting point. The design-token-extractor skill inventories the values already in your product and proposes a named set; the /design-tokens command runs the same pass inside a repository; and maintain a design system with Claude Code covers keeping the set honest as the product grows. Tokens are also what make design-to-code produce a diff a developer will merge rather than a rewrite. The wider workflow is in the Claude Design guide.

Frequently asked questions

What counts as a design token?
Any design decision worth naming: palette colors, spacing steps, type sizes and weights, border radii, shadows, motion durations, and breakpoints. Teams usually split them into primitive tokens, which are raw values like a hex code, and semantic tokens, which describe intent, like surface or danger, and point at the primitives.
Why do tokens matter more now than they did?
Because generators read them. Tools that build interfaces from a prompt reuse whatever system you give them, so a clean token set produces output that looks like your product and a messy one produces plausible-looking output in someone else's house style. Tokens turned into the interface between your brand and every AI tool that touches it.
How do I start if we have no tokens at all?
Extract before you design. Have an agent inventory every color, spacing value, font size, and radius in the codebase, group the near-duplicates, and propose a named scale. You will usually find a handful of intentional values buried under dozens of accidental ones, and the grouping is the real work.

Related